Lintel on GitHub Copilot. Generated from skills/jobs/SKILL.md; edit the canonical file, then run
li-copilot init.
- Resource root:
../../.. from this skill's base directory (the Lintel source with bin/,
lib/, skills/). Write plans, state and evidence into the working repository's .claude/
tree, never into the resource root.
- Skill-relative paths: this skill's own
scripts/, references/ and data/ folders (and a
<base> that the workflow defines as its own directory) mean
../../../skills/jobs/ in the Lintel source, not this generated folder.
${LINTEL_SKILLS_DIR:-skills} means the skills root, ../../../skills. A bin/li-run step
runs in the working repository, so use $LINTEL_SKILLS_DIR/jobs/ there.
- Shell steps: run Bash snippets with Bash (Git for Windows'
bash.exe on Windows, never
System32\bash.exe). Save a snippet to a temporary .sh file and run
bash "<resource root>/bin/li-run" <file>; it prepares LINTEL_SOURCE_ROOT, LINTEL_REPO_ROOT
and the profile context.
- Tools: Read=
view, Write=create, Edit=edit, Bash=bash/powershell, Grep=grep,
Glob=glob, AskUserQuestion=ask_user, TodoWrite=the plan checklist, Task or a named role=task
with that custom agent, WebFetch=web_fetch.
- Other Lintel workflows are native skills: invoke
/li-<name> rather than reading their
files. Named roles such as CodeReviewer are custom agents.
You are the jobs skill — lifecycle controller for in-flight Lintel curated flows (v3.8 Feature 1).
What this skill does
Provides 5 subcommands to operate on .claude/runtime/jobs/ (the repo-local job store):
list (default) — read actual selected repository job records without regenerating the derived view
continue <id> — resume the job at its current_step (delegates to /li-resume mechanic)
replan <id> — re-run the whole job or pick which steps to redo (supports build-on, not just start-over)
abort <id> — move job to _archive/ with status: aborted, apply discard policy, skip keep-promotion
branch <id> — create a parallel job from the same starting point (new job-id, copies job.yaml + outputs/)
Mutation helpers regenerate their derived _active.md job view; list reads actual job records without
regeneration. Missing records are unobserved, not proof that no mapped work is open.
Repository jobs do not select a cross-repository registry or personal profile.
Set LINTEL_JOBS_REGISTRY before sourcing the helper only for an explicitly
selected, separately authorized registry destination. Without that selection,
repository mutations update only their local view; existing global records are
neither scanned nor rewritten. A path setting selects the destination, not host
permission. The legacy global job-store layout remains available for its
explicitly selected scope.
Jobs auto-spawn/hooks remain dormant unless separately activated and verified.
Apply task-relevant intake: no fixed question API
and no automatic choice when a material decision is unresolved.
When to use
- "Which job records are open?" →
list; /li-status also inspects selected mapped/cyclic work
- "I want to pick up that customer-engagement cycle from Tuesday" →
continue <id>
- "Restart this plan but keep the discovery output" →
replan <id> with partial-redo prompt
- "Kill this. I started wrong." →
abort <id>
- "I want to try the same design with motion=kinetic instead of moderate" →
branch <id>
When NOT to use
- Mid-task code-editing — jobs is lifecycle, not editing
- For non-workflow_root skills — single-shot diagnostics such as
/li-doctor don't spawn jobs
- To enforce ordering between steps — that's
job.yaml.blocked_until, evaluated by job_can_start <id> <step> in bin/_jobs.sh, not this skill
Inputs
list — no args
continue <id> — <id> required (validates against jobs//)
replan <id> — <id> required; optional --from <step> to restart from specific phase
abort <id> — <id> required; optional --reason "<text>" (audit-logged)
branch <id> — <id> required; optional --name <suffix> (default: -branch-)
Workflow
Step 1 — Parse subcommand
sub="${1:-list}"
job_id="${2:-}"
case "$sub" in
list|continue|replan|abort|branch) ;;
*) echo "Usage: /li-jobs <list|continue|replan|abort|branch> [<id>]"; exit 2 ;;
esac
Step 2 — Source helper
BIN_DIR="${LINTEL_SOURCE_ROOT:?select trusted source}/bin"
export LINTEL_JOBS_NO_INIT=1
source "$BIN_DIR/_jobs.sh"
Step 3 — Dispatch
list:
list_jobs --read-only
stale_jobs 24
continue <id>:
- Verify
.claude/runtime/jobs/<id>/job.yaml exists.
- Determine the resume target. For a
tree-schema plan, call
job_resume_point <id> (deepest incomplete + startable WBS node-path, e.g.
1.1.a); for flat/phased plans, read current_step from job.yaml. See
/li-resume Step 2.5.
For mapped work the original tasks and verified lifecycle reference remain
authoritative; job steps are execution observations, not a second backlog.
Retain blocked and unknown-age work in the report even when it is not startable.
- Invoke
/li-resume --job <id> (resume skill reads from job dir, not loose .claude/runtime/state/00-state.md).
- Update last_touched.
replan <id>:
- Verify job exists.
- ask_user: "Replan whole job, OR pick a step to restart from?"
- Whole → preserve existing history and original map/IDs, then select an explicit
retry/loop-back through
/li-cycle; do not erase evidence to simulate a fresh start
- From step → operator picks;
/li-resume --from <step> style
- Update last_touched.
abort <id>:
- Verify job exists.
- ask_user confirm + optional
--reason.
- Call
job_archive <id> ABORTED.
- Surface
[lintel] Job <id> aborted, archived to _archive/<date>/<id>/.
branch <id>:
- Verify source job exists.
- Use
job_create for a distinct record; never copy the old job_id into a new folder.
- Retain the original selected map as a reference. A new initiative needs explicit
map selection, not copied approval/status or a directory name stripped from the job ID.
- Copy only authorized inputs/output candidates, preserving source history and conflicts.
They are candidates, not independently reviewed completion.
- Use
job_update for the new current step and verify actual persistence before
reporting both records. There is no autonomous branch scheduler.
Step 4 — Audit
The helpers in bin/_jobs.sh record their own operations in the jobs category
(.claude/runtime/audit/jobs.jsonl on the v5 layout): job_begin (job_create),
job_set_steps, job_update and job_end (job_archive). job_stale_warn is recorded by the
dormant job-stale-warn hook, not by bin/_jobs.sh. This skill emits no separate job_action
record; continue, replan and branch leave only the helper records they cause. These audit
lines are advisory, while a failed job.yaml write makes the helper return non-zero:
{"ts":"...","kind":"job_update","operator":"...","cycle_id":"...","job_id":"...","step":"PLAN","status":"IN_PROGRESS"}
Voice tier behavior
voice: internal. No customer-bound output. Operator-only.
Status protocol
- DONE — the selected operation completed;
list does not regenerate any view,
while mutations report their actual local-view and selected-registry outcomes
- DONE_WITH_CONCERNS — operation succeeded with warnings (e.g., source job partially missing files; branch operation proceeded with caveats)
- BLOCKED — subcommand requires
<id> but not provided, OR job doesn't exist
- NEEDS_CONTEXT —
replan asked operator to pick scope but no answer
Pause-points
abort requires ask_user confirm (destructive — moves files out of working set)
replan whole confirms the retry/loop-back scope while preserving original
artifacts, map/IDs and review/voice provenance; retain existing authorization
branch proceeds without confirm (additive, non-destructive)
Hop-in support
YES — solo-invocable. Designed to be called anytime.
Integration
Reads:
.claude/runtime/jobs/_active.md (repo-local active list)
- An explicitly selected and separately authorized cross-repo registry, when requested;
list does not read it implicitly
.claude/runtime/jobs/<id>/job.yaml
.claude/runtime/jobs/<id>/00-state.md
- Supplied job metadata and already verified mode/context; no implicit personal-profile scan
Writes:
.claude/runtime/jobs/<id>/job.yaml (last_touched updates)
.claude/runtime/jobs/_active.md (regenerated)
- The separately authorized
LINTEL_JOBS_REGISTRY destination, only when explicitly selected
.claude/runtime/jobs/_archive/<date>/<id>/ (on abort)
.claude/runtime/audit/jobs.jsonl
Calls into:
bin/_jobs.sh helper (sourced) — incl. job_resume_point (node-path), job_can_start (blocked_until), job_set_steps / job_step_status
/li-resume (for continue subcommand)
/li-cycle (for replan whole)
Optional hooks (dormant by default):
job-begin, job-end and job-stale-warn are not activated by this skill.
- File presence is not registration/execution. Do not use dormant promotion recipes
to copy failed/colliding outputs over original mapped artifacts.
Anti-patterns
- Direct edit of
_active.md — it's regenerated. Edit job.yaml instead.
- Manual
rm -rf jobs/<id>/ — bypasses cleanup-policy + audit. Use abort.
- Branching to keep multiple "active" versions of same work — branch is for divergent exploration, not version-control.
- Erasing provenance during a replan — preserve the original voice/review
evidence and mark changed-content checks stale; branch only for an actual divergent path.
Failure recovery
<id> not found: BLOCKED with diagnostic
_active.md corrupt: regenerate via helper
job.yaml malformed: surface diff, offer manual repair
- Operation succeeds but
_active.md regeneration fails: warn, continue (non-critical)
Recommended next steps after invocation
- After
list: pick an id, run continue / abort / branch
- After
continue: work proceeds in the job's current_step
- After
abort: review _archive/<date>/<id>/job.yaml for forensics
- After
branch: both jobs visible in next list
- After
replan: cycle continues from chosen step
See also
/li-status (read-only selected-work/cycle inspection with supplementary job observations)
/li-resume (the underlying state-recovery mechanic)
/li-cycle (the most common workflow_root skill)
/li-plan (also workflow_root post-v3.8 Feature 2.1)
docs/concepts/jobs-system.md (architecture doc)
Cycle-position footer
Close your report with the shared position footer. Outside an active cycle it renders the thin
ambient line; inside one it shows the operator's position + next step:
source "${LINTEL_SOURCE_ROOT:?select trusted source}/lib/cycle-footer.sh"
render_cycle_footer # auto: thin when no cycle, full/--compact when in one
See ADR-0003.
1---2name: li-jobs3description: Use when listing or steering repository-local jobs; job lifecycle observations do not replace selected work-map authority or establish all open work.4metadata:5 source: https://skillmd.com/skills/jokerman89/li-jobs6---78> **Lintel on GitHub Copilot.** Generated from `skills/jobs/SKILL.md`; edit the canonical file, then run9> `li-copilot init`.10> - **Resource root:** `../../..` from this skill's base directory (the Lintel source with `bin/`,11> `lib/`, `skills/`). Write plans, state and evidence into the working repository's `.claude/`12> tree, never into the resource root.13> - **Skill-relative paths:** this skill's own `scripts/`, `references/` and `data/` folders (and a14> `<base>` that the workflow defines as its own directory) mean15> `../../../skills/jobs/` in the Lintel source, not this generated folder.16> `${LINTEL_SKILLS_DIR:-skills}` means the skills root, `../../../skills`. A `bin/li-run` step17> runs in the working repository, so use `$LINTEL_SKILLS_DIR/jobs/` there.18> - **Shell steps:** run Bash snippets with Bash (Git for Windows' `bash.exe` on Windows, never19> `System32\bash.exe`). Save a snippet to a temporary `.sh` file and run20> `bash "<resource root>/bin/li-run" <file>`; it prepares `LINTEL_SOURCE_ROOT`, `LINTEL_REPO_ROOT`21> and the profile context.22> - **Tools:** Read=`view`, Write=`create`, Edit=`edit`, Bash=`bash`/`powershell`, Grep=`grep`,23> Glob=`glob`, AskUserQuestion=`ask_user`, TodoWrite=the plan checklist, Task or a named role=`task`24> with that custom agent, WebFetch=`web_fetch`.25> - **Other Lintel workflows** are native skills: invoke `/li-<name>` rather than reading their26> files. Named roles such as `CodeReviewer` are custom agents.2728You are the `jobs` skill — lifecycle controller for in-flight Lintel curated flows (v3.8 Feature 1).2930## What this skill does3132Provides 5 subcommands to operate on `.claude/runtime/jobs/` (the repo-local job store):3334- `list` (default) — read actual selected repository job records without regenerating the derived view35- `continue <id>` — resume the job at its `current_step` (delegates to `/li-resume` mechanic)36- `replan <id>` — re-run the whole job or pick which steps to redo (supports build-on, not just start-over)37- `abort <id>` — move job to `_archive/` with `status: aborted`, apply discard policy, skip keep-promotion38- `branch <id>` — create a parallel job from the same starting point (new job-id, copies job.yaml + outputs/)3940Mutation helpers regenerate their derived `_active.md` job view; `list` reads actual job records without41regeneration. Missing records are unobserved, not proof that no mapped work is open.42Repository jobs do not select a cross-repository registry or personal profile.43Set `LINTEL_JOBS_REGISTRY` before sourcing the helper only for an explicitly44selected, separately authorized registry destination. Without that selection,45repository mutations update only their local view; existing global records are46neither scanned nor rewritten. A path setting selects the destination, not host47permission. The legacy global job-store layout remains available for its48explicitly selected scope.49Jobs auto-spawn/hooks remain dormant unless separately activated and verified.50Apply [task-relevant intake](../../../skills/define/references/intake.md): no fixed question API51and no automatic choice when a material decision is unresolved.5253## When to use5455- "Which job records are open?" → `list`; `/li-status` also inspects selected mapped/cyclic work56- "I want to pick up that customer-engagement cycle from Tuesday" → `continue <id>`57- "Restart this plan but keep the discovery output" → `replan <id>` with partial-redo prompt58- "Kill this. I started wrong." → `abort <id>`59- "I want to try the same design with motion=kinetic instead of moderate" → `branch <id>`6061## When NOT to use6263- Mid-task code-editing — jobs is lifecycle, not editing64- For non-workflow_root skills — single-shot diagnostics such as `/li-doctor` don't spawn jobs65- To enforce ordering between steps — that's `job.yaml.blocked_until`, evaluated by `job_can_start <id> <step>` in `bin/_jobs.sh`, not this skill6667## Inputs6869**`list`** — no args70**`continue <id>`** — `<id>` required (validates against jobs/<id>/)71**`replan <id>`** — `<id>` required; optional `--from <step>` to restart from specific phase72**`abort <id>`** — `<id>` required; optional `--reason "<text>"` (audit-logged)73**`branch <id>`** — `<id>` required; optional `--name <suffix>` (default: <orig>-branch-<N>)7475## Workflow7677### Step 1 — Parse subcommand7879```bash80sub="${1:-list}"81job_id="${2:-}"82case "$sub" in83 list|continue|replan|abort|branch) ;;84 *) echo "Usage: /li-jobs <list|continue|replan|abort|branch> [<id>]"; exit 2 ;;85esac86```8788### Step 2 — Source helper8990```bash91BIN_DIR="${LINTEL_SOURCE_ROOT:?select trusted source}/bin"92export LINTEL_JOBS_NO_INIT=193source "$BIN_DIR/_jobs.sh"94```9596### Step 3 — Dispatch9798**`list`:**99```bash100list_jobs --read-only101stale_jobs 24102```103104**`continue <id>`:**1051. Verify `.claude/runtime/jobs/<id>/job.yaml` exists.1062. Determine the resume target. For a `tree`-schema plan, call107 `job_resume_point <id>` (deepest incomplete + startable WBS node-path, e.g.108 `1.1.a`); for flat/phased plans, read `current_step` from job.yaml. See109 `/li-resume` Step 2.5.110 For mapped work the original tasks and verified lifecycle reference remain111 authoritative; job steps are execution observations, not a second backlog.112 Retain blocked and unknown-age work in the report even when it is not startable.1133. Invoke `/li-resume --job <id>` (resume skill reads from job dir, not loose `.claude/runtime/state/00-state.md`).1144. Update last_touched.115116**`replan <id>`:**1171. Verify job exists.1182. ask_user: "Replan whole job, OR pick a step to restart from?"119 - Whole → preserve existing history and original map/IDs, then select an explicit120 retry/loop-back through `/li-cycle`; do not erase evidence to simulate a fresh start121 - From step → operator picks; `/li-resume --from <step>` style1223. Update last_touched.123124**`abort <id>`:**1251. Verify job exists.1262. ask_user confirm + optional `--reason`.1273. Call `job_archive <id> ABORTED`.1284. Surface `[lintel] Job <id> aborted, archived to _archive/<date>/<id>/`.129130**`branch <id>`:**1311. Verify source job exists.1322. Use `job_create` for a distinct record; never copy the old `job_id` into a new folder.1333. Retain the original selected map as a reference. A new initiative needs explicit134 map selection, not copied approval/status or a directory name stripped from the job ID.1354. Copy only authorized inputs/output candidates, preserving source history and conflicts.136 They are candidates, not independently reviewed completion.1375. Use `job_update` for the new current step and verify actual persistence before138 reporting both records. There is no autonomous branch scheduler.139140### Step 4 — Audit141142The helpers in `bin/_jobs.sh` record their own operations in the `jobs` category143(`.claude/runtime/audit/jobs.jsonl` on the v5 layout): `job_begin` (`job_create`),144`job_set_steps`, `job_update` and `job_end` (`job_archive`). `job_stale_warn` is recorded by the145dormant `job-stale-warn` hook, not by `bin/_jobs.sh`. This skill emits no separate `job_action`146record; `continue`, `replan` and `branch` leave only the helper records they cause. These audit147lines are advisory, while a failed `job.yaml` write makes the helper return non-zero:148149```json150{"ts":"...","kind":"job_update","operator":"...","cycle_id":"...","job_id":"...","step":"PLAN","status":"IN_PROGRESS"}151```152153## Voice tier behavior154155`voice: internal`. No customer-bound output. Operator-only.156157## Status protocol158159- **DONE** — the selected operation completed; `list` does not regenerate any view,160 while mutations report their actual local-view and selected-registry outcomes161- **DONE_WITH_CONCERNS** — operation succeeded with warnings (e.g., source job partially missing files; branch operation proceeded with caveats)162- **BLOCKED** — subcommand requires `<id>` but not provided, OR job doesn't exist163- **NEEDS_CONTEXT** — `replan` asked operator to pick scope but no answer164165## Pause-points166167- `abort` requires ask_user confirm (destructive — moves files out of working set)168- `replan whole` confirms the retry/loop-back scope while preserving original169 artifacts, map/IDs and review/voice provenance; retain existing authorization170- `branch` proceeds without confirm (additive, non-destructive)171172## Hop-in support173174YES — solo-invocable. Designed to be called anytime.175176## Integration177178**Reads:**179- `.claude/runtime/jobs/_active.md` (repo-local active list)180- An explicitly selected and separately authorized cross-repo registry, when requested;181 `list` does not read it implicitly182- `.claude/runtime/jobs/<id>/job.yaml`183- `.claude/runtime/jobs/<id>/00-state.md`184- Supplied job metadata and already verified mode/context; no implicit personal-profile scan185186**Writes:**187- `.claude/runtime/jobs/<id>/job.yaml` (last_touched updates)188- `.claude/runtime/jobs/_active.md` (regenerated)189- The separately authorized `LINTEL_JOBS_REGISTRY` destination, only when explicitly selected190- `.claude/runtime/jobs/_archive/<date>/<id>/` (on abort)191- `.claude/runtime/audit/jobs.jsonl`192193**Calls into:**194- `bin/_jobs.sh` helper (sourced) — incl. `job_resume_point` (node-path), `job_can_start` (blocked_until), `job_set_steps` / `job_step_status`195- `/li-resume` (for `continue` subcommand)196- `/li-cycle` (for `replan whole`)197198**Optional hooks (dormant by default):**199- `job-begin`, `job-end` and `job-stale-warn` are not activated by this skill.200- File presence is not registration/execution. Do not use dormant promotion recipes201 to copy failed/colliding outputs over original mapped artifacts.202203## Anti-patterns204205- **Direct edit of `_active.md`** — it's regenerated. Edit job.yaml instead.206- **Manual `rm -rf jobs/<id>/`** — bypasses cleanup-policy + audit. Use `abort`.207- **Branching to keep multiple "active" versions of same work** — branch is for divergent exploration, not version-control.208- **Erasing provenance during a replan** — preserve the original voice/review209 evidence and mark changed-content checks stale; branch only for an actual divergent path.210211## Failure recovery212213- `<id>` not found: BLOCKED with diagnostic214- `_active.md` corrupt: regenerate via helper215- `job.yaml` malformed: surface diff, offer manual repair216- Operation succeeds but `_active.md` regeneration fails: warn, continue (non-critical)217218## Recommended next steps after invocation219220- After `list`: pick an id, run `continue` / `abort` / `branch`221- After `continue`: work proceeds in the job's current_step222- After `abort`: review `_archive/<date>/<id>/job.yaml` for forensics223- After `branch`: both jobs visible in next `list`224- After `replan`: cycle continues from chosen step225226## See also227228- `/li-status` (read-only selected-work/cycle inspection with supplementary job observations)229- `/li-resume` (the underlying state-recovery mechanic)230- `/li-cycle` (the most common workflow_root skill)231- `/li-plan` (also workflow_root post-v3.8 Feature 2.1)232- `docs/concepts/jobs-system.md` (architecture doc)233234## Cycle-position footer235236Close your report with the shared position footer. Outside an active cycle it renders the thin237ambient line; inside one it shows the operator's position + next step:238239```bash240source "${LINTEL_SOURCE_ROOT:?select trusted source}/lib/cycle-footer.sh"241render_cycle_footer # auto: thin when no cycle, full/--compact when in one242```243244See [ADR-0003](../../../.claude/decisions/0003-cycle-position-footer.md).